Tomas Grosup T-Gro

Expert Reviewer Extraction Pipeline

Generate folder-scoped instructions, topic-scoped skills, and a multi-dimensional review agent from a GitHub user's public review history. Produces anonymized, deduplicated, Copilot-compatible .github/ artifacts.

Scale Warning

This pipeline processes thousands of GitHub items (typically 3,000–10,000+ issues, PRs, discussions, and review comments spanning a decade). It will not fit in a single context window.

Use sub-agents for everything. The orchestrator manages SQLite state and dispatches work. Sub-agents do the heavy lifting:

Collection: one sub-agent per repo × date range chunk. Parallelize aggressively (6+ concurrent agents).

	#!/usr/bin/env -S deno run --allow-read --allow-write --allow-env
	import { Server } from "npm:@modelcontextprotocol/sdk@0.5.0/server/index.js";
	import { StdioServerTransport } from "npm:@modelcontextprotocol/sdk@0.5.0/server/stdio.js";
	import {
	CallToolRequestSchema,
	ListToolsRequestSchema,
	} from "npm:@modelcontextprotocol/sdk@0.5.0/types.js";

	const server = new Server({
	name: "ffs-instructions",